Third Prototype
Let’s see our third prototype! There are several features added for this version.
- Cross-fading:
- We changed the approach to switch the clips from jump-cut to cross-fading. Whenever we switch the clips to play, the system fades out the playing video first, then fade-in the following clip.
- Cross-fading is a smart approach, but it’s not magic. To make seamless switching, there are several tips to be aware of. We ask actors to always be in the same pose for starting and ending of each response. For example, they can use their hands to reinforce their actions during the conversation, but their hands must be back to the lap at the end. We also need to use some line or marker tapes to remind actors/actresses of the position/rotation of their bodies.
- Not always recording/listening (for the Chatbot)
- In the beginning, we keep listening and recording what the player says because we want to know if the player interrupts the patient.
- Later, we figure out it is trembling ourselves. There are too many interactions detected as interruptions in a natural conversation. If you pause too long between your sentences, our chatbot will send them as two separated inputs. If you unconsciously acknowledge when the patient telling her story, our chatbot will stop the clips and send the player’s acknowledgment as new inputs.
- Noise Cancelling
- The noise-canceling technique could be used for sound input and output. For the output, which means the noise in the clips, there are two ways we tried to reduce ambiance sounds. Choosing an external microphone like an over-ear or short-gun might be helpful. The directional microphone could reduce the noise and perceive the clear voice from patients. But no matter which microphone we use, it’s impossible to record the clips without noise. Therefore, all the clips require to be post-process in the studio. We use the Premiere to eliminate the noise, and it works out pretty well.